Method and apparatus for generating conversation, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-2021280189-A1
Application number	US-202117326917-A
Country	US
Kind code	A1
Filing date	May 21, 2021
Priority date	Sep 18, 2020
Publication date	Sep 9, 2021
Grant date	—

Abstract

A first expression corresponding to an input statement is obtained from a conversational graph. The conversational graph includes expressions having association relationships therebetween and conversation target clusters having association relationships therebetween. Each conversation target cluster includes at least two expressions. A second expression associated with the first expression is obtained from the conversational graph based on the association relationships of expressions and the association relationships of conversational target clusters in the conversational graph. A reply statement is generated based on the second expression and the input statement.

First claim

What is claimed is:

1 . A method for generating a conversation, comprising:
obtaining an input statement; obtaining a first expression corresponding to the input statement from a conversational graph, the conversational graph comprising expressions having association relationships therebetween and conversation target clusters having association relationships therebetween, and each conversation target cluster comprising at least two expressions; obtaining a second expression associated with the first expression from the conversational graph based on the association relationships of expressions and the association relationships of conversational target clusters in the conversational graph; and generating a reply statement based on the second expression and the input statement.

2 . The method of claim 1 , wherein obtaining the second expression comprises:
obtaining first candidate expressions associated with the first expression from the conversational graph based on the association relationships of expressions in the conversational graph; determining a second conversation target cluster based on an association relationship between a first conversation target cluster to which the first expression belongs and a respective candidate conversation target cluster to which each first candidate expression belongs; and obtaining the second expression associated with the first expression from the second conversation target cluster.

3 . The method of claim 2 , wherein the association relationships of expressions comprise directions of edges connecting the expressions and first weights of respective edges, and
wherein, obtaining the first candidate expressions comprises:
obtaining the first candidate expressions from the conversational graph based on the directions and the first weights, each first candidate expression corresponding to an edge pointing from the first expression to the first candidate expression, and the first weight of the edge being greater than a first threshold.

4 . The method of claim 2 , wherein the association relationships of expressions comprise directions of edges connecting the expressions, and
wherein obtaining the first candidate expressions comprises:
obtaining the first candidate expression based on the directions and the number of edges, the direction of the edge between each first candidate expression and the first expression being from the first expression to the first candidate expression, and the number of edges corresponding to each first candidate expression being greater than a second threshold.

5 . The method of claim 2 , wherein obtaining the first candidate expressions comprises:
obtaining second candidate expressions associated with the first expression from the conversational graph based on the association relationships of expressions; and obtaining the first candidate expressions from the second candidate expressions based on the number of times of selecting each second candidate expression.

6 . The method of claim 2 , wherein determining the second conversation target cluster comprises:
determining that the second conversation target cluster is the same as the first conversation target cluster based on a candidate conversation target cluster to which a first candidate expression belongs being the first conversation target cluster.

7 . The method of claim 2 , wherein the association relationships of conversation target clusters comprise directions of edges connecting the conversation target clusters and second weights of respective edges, and
wherein determining the second conversation target cluster comprises:
obtaining the second conversation target cluster from the candidate conversation target clusters based on the direction of the edge connecting the first conversation target cluster and each candidate conversation target cluster and the second weight of the edge, the edge pointing from the first conversation target cluster to the second conversation target cluster, and the second weight of the edge being greater than a third threshold.

8 . The method of claim 2 , wherein the association relationships of conversation target clusters comprise directions of edges connecting the conversation target clusters and second weights of respective edges, each conversation target cluster corresponding to a conversation target, and
wherein determining the second conversation target cluster comprises:
determining a to-be-guided conversation target by recognizing the input statement;
determining a third conversation target cluster based on a matching degree between the to-be-guided conversation target and each conversation target cluster in the conversational graph; and
obtaining the second conversation target cluster from the candidate conversation target clusters based on the direction of the edge between each candidate conversation target cluster and the third conversation target cluster and the second weight of the edge.

9 . The method of claim 1 , wherein obtaining the first expression comprises:
obtaining word segmentations of the input statement by performing word segmentation processing on the input statement; and determining the first expression from the conversational graph based on an association degree between each word segmentation and each expression contained in the conversational graph.

10 . The method of claim 9 , wherein determining the first expression based on the association degree comprises:
obtaining third candidate expressions from the conversational graph based on the association degree; and obtaining the first expression from the third candidate expressions based on a similarity degree between each third candidate expression and the input statement.

11 . The method of claim 1 , further comprising:
obtaining a conversational data set, the conversational data set comprising groups of conversation; extracting expressions from the conversational data set; determining the association relationships between the expressions based on co-occurrence ways of every two expressions in a common group of conversation; and determining the conversation target clusters and the association relationships of conversation target clusters by clustering the expressions based on the association relationships of expressions.

12 . The method of claim 11 , wherein extracting the expressions comprises:
obtaining a syntax tree corresponding to each conversation contained in the conversational data set by parsing the conversation; and extracting an expression corresponding to each conversation from a root node of the syntax tree corresponding to the conversation.

13 . An electronic device, comprising:
at least one processor; and a memory, communicatively coupled to the at least one processor, wherein the memory is configured to store instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is configured to: obtain an input statement; obtain a first expression corresponding to the input statement from a conversational graph, the conversational graph comprising expressions having association relationships therebetween and conversation target clusters having association relationships therebetween, and each conversation target cluster comprising at least two expressions; obtain a second expression associated with the first expression from the conversational graph based on the association relationships of expressions and the association relationships of conversational target clusters in the conversational graph; and generate a reply statement based on the second expression

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Xu Jun
• Lei Zeyang
• Niu Zhengyu
• Wu Hua
• Wang Haifeng

Classifications

• G06F16/367

  Ontology · CPC title

• G06F16/9024

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F16/3329 Primary

  Natural language query formulation · CPC title

• G10L15/083

  Recognition networks (G10L15/142, G10L15/16 take precedence) · CPC title

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title